All Research
Page 8 of 10 · 219 reports
Good asset, bad price: net cash on the balance sheet and a second curve forming in energy storage, yet 404 dollars already prepays far too much for the long-dated Robotaxi/FSD/Optimus narrative, leaving no margin of safety. Rating Watch: a verifiable intrinsic value of roughly 90 to 160 dollars sits well below today's price, so the question is no longer how thick the margin is but that it barely exists.
A small AI infrastructure company running two businesses, GPU cloud plus AI/HPC data center colocation, with revenue scaling fast but capex scaling faster, deeply negative FCF, heavy customer concentration, and majority control by parent Bit Digital; at roughly 11.6x P/S there is no margin of safety. Rating Watch: a high-uncertainty growth infrastructure name to track rather than own at today's price.
2025 revenue of RMB 37.12 billion and net profit of RMB 12.78 billion, with overseas at 44% of sales and THE MONSTERS a single IP contributing 38.1% of revenue; but at the current HK$152.90, against a conservative owner-earnings base of RMB 8.9 billion the P/OE sits near 20x, in the lower half of the neutral range, so the price is not cheap. The core risks are the lifecycle of a blockbuster IP and the reversion of margins. Rating Watch: a great business priced for proven success rather than a wide margin of safety.
A high-quality, asset-light platform with strong cash generation, but single-segment reporting leaves Temu's economics opaque, while the VIE structure and cross-border regulation layer on discounts. At a conservative 10x Owner Earnings the valuation looks reasonable rather than cheap, with no clear margin of safety; fair range is $95-125 and the current $94.97 sits near the lower bound. Rating Watch: a strong cash machine whose structural complexity and price still argue for watching rather than committing heavily.
A long-term, platform-monopoly stickiness machine with relentless cash generation and a front-row reinvestment seat into the AI era; at the current price the market cap sits near $3.15 trillion, putting it on roughly 28x conservative Owner Earnings and a strict 43x FCF. At $423.54 the stock is close to fair value but lacks a margin of safety, with a sensible buy range of $300–360 (fair value $390–470 at a 20–30% discount). Rating Watch: a superb business at a reasonable-to-slightly-rich price, worth following closely rather than chasing here.
Built on AI custom silicon, optical interconnect, and data-center networking, with the data center making up three quarters of FY2026 revenue and genuine AI infrastructure demand behind it; yet today's 104x P/FCF and a 0.96% FCF yield (far below the 4.59% 10-year Treasury) already prepay the optimistic case. Fair value sits at $60-90, and the margin of safety is thin. Rating Watch: a strengthening AI-infrastructure platform whose price has run ahead of its value.
The world's strongest compound platform of attention distribution, performance advertising, and the social graph, with 2025 revenue of 201 billion dollars, operating cash flow of 115.8 billion dollars, and FCF of 43.6 billion dollars; yet 2026 capex guidance runs 125 billion to 145 billion dollars. At today's 611.21 dollars, the stock trades at roughly 35x Owner Earnings, leaving no clear margin of safety. Rating Watch: a superb business whose current price prepays for a decade of high-quality growth rather than buying it at a discount; fair buy 330 to 450 dollars.
A top-tier compounding platform (retail network + AWS + advertising + third-party sellers), with 2025 revenue of 716.9 billion dollars and operating profit of 80.0 billion dollars; but AI capex has surged to 128.3 billion dollars, leaving TTM free cash flow at just 1.2 billion dollars. At the current 264.86 dollars, the stock sits near the upper edge of the optimistic case, and the margin of safety is thin. Fair buy 150-210 dollars, comfortable entry 120-155 dollars. Rating Watch: a world-class business, but priced as expensive excellence rather than cheap greatness.
AI cybersecurity has graduated from a "product feature" into the control plane and governance layer of the AI value chain, spanning models, agents, data, identity, runtime, development, and the SOC. Two budget curves diverge: using AI to do security (replacing or upgrading existing SIEM/SOAR/XDR/MDR) lands fast in the near term, while protecting AI itself (AI-SPM/agent identity/RAG permissions/MCP governance/AI gateway) carries greater long-run elasticity. The public companies with the clearest revenue leverage are Palo Alto Networks (NGS ARR $4.8 billion, +37%; RPO $13 billion), CrowdStrike (FY25 ARR $4.24 billion), Fortinet (Unified SASE/SecOps ARR +26%/+30%), Zscaler, and Rubrik (Subscription ARR $1.09 billion, +39%); the defensive beneficiaries are Microsoft, Datadog, Cloudflare, Check Point, and Qualys. The most bubble-prone are pure prompt injection, LLM firewalls, and single-point red-teaming, most of which will be absorbed or acquired by the platforms. Rating Watch: a sector-level control plane in formation, where business certainty and valuation attractiveness have already separated.
The profit pool in AI industrial manufacturing first lands with companies that already have an installed base, workflow entry points, and strong engineering-delivery capability (Siemens / Schneider / Rockwell / ABB / Emerson / Honeywell / PTC / Dassault / KLA / AMAT / Cognex / Keyence / Teradyne). What monetizes first is semiconductor defect detection, industrial-vision quality inspection, predictive maintenance, digital twins, and robot simulation. Humanoid robots and cross-plant autonomous control remain conceptual. Rating Watch: this is a real-cash-flow, delivery-dependent theme where the durable winners are the platform vendors and equipment makers, not the pure-narrative names.
In AI retail and e-commerce, the first thing to monetize is not "chat" but the advertising and transaction loop wrapped around shopping traffic. Retail media remains the most mature, highest-margin profit pool: Amazon Ads totaled $68.635 billion for full-year 2025, Walmart global advertising rose 37% in Q4 FY26 (U.S. Walmart Connect +41%), DoorDash advertising runs above a $1 billion annualized run rate across 150,000+ advertisers, and eBay advertising reached $581 million in Q1 2026 (2.6% of GMV). Shopping agents are moving from "product launch" to "partial transaction loop": Amazon says its shopping AI assistant helped 300 million+ customers in 2025, Shopify's Agentic Storefronts already connect to ChatGPT, Google AI Mode, and Copilot, and PayPal is turning payments, identity, and risk control into an agent layer. The profit pools accrue mainly to three groups: platform winners (Amazon, Walmart, Shopify, Alibaba, JD, MercadoLibre, Instacart, DoorDash), the settlement/authorization layer (PayPal), and the picks-and-shovels vendors (Criteo, Constructor, Gorgias). Dynamic pricing is squeezed by FTC surveillance-pricing scrutiny; Instacart halted platform item price tests in December 2025. Gartner expects that by 2027 more than 40% of agentic AI projects may be cancelled over unclear cost and value. Rating Overweight: the durable profit pools sit with transaction-owning platforms, the settlement layer, and data-rich tooling, not with standalone AI gadgets.
Over the next two to three years the AI gateway will settle into a two-layer control structure: the bottom layer (OS, browser, search, super-apps) owns distribution and permissions, while the top layer (AI assistants and agent platforms) fights for the right to execute user intent. The surest money still sits in enterprise AI gateways, AI search advertising, and cloud inference; pure consumer agent companies still monetize more weakly than Big Tech's system-level moats. Rating Watch: a real but unevenly realized opportunity where entry rights, not model quality, decide the winners.
AI in supply chain and logistics has moved past the demo stage into real budget line items; the first to pay off are closed-loop, quantifiable-ROI use cases—demand forecasting, WMS/TMS, freight visibility, warehouse robotics/ASRS, and parcel sortation. The profit pool sits mostly with the vertical platforms closest to the workflow (Kinaxis / Manhattan / Descartes / WiseTech / AutoStore / Kardex / Symbotic), while logistics carriers look more like margin beneficiaries of AI than software-revenue beneficiaries. Rating Watch: profit accrues to the platforms that own the workflow, the data, and the customer relationship, not to the AI narrative itself.
As AI agents go mainstream, identity security is being elevated into a unified "identity control plane" that governs humans, workloads, service accounts, API keys, OAuth apps, certificates, tokens, CI/CD, bots, and agents alike. Non-Human Identity (NHI) is becoming a new budget entry point—the MCP spec already mandates the resource parameter and token audience validation while prohibiting token passthrough. The first budget pools to land are PAM, machine identity, secrets, and IGA-CIEM extensions, because they already exist and are strongly compliance-driven. Direct beneficiaries: SailPoint (FY26 ARR $1.125 billion, SaaS ARR +38%), CyberArk, Okta (FY26 Q4 RPO +15%), Microsoft Entra, IBM/HashiCorp Vault, and JFrog (Security Core already 10% of ARR); CrowdStrike's acquisition of SGNL and Cisco's planned acquisition of Astrix confirm that the NHI control plane is the next-generation entry point. Cloud providers' built-in IAM will keep commoditizing basic secrets and CIEM, while cross-cloud, cross-SaaS ownership, least-privilege, and runtime enforcement remain the profit pool. Rating Watch: identity security is on track to become the control plane of the AI agent era, but budgets land first in mature control points before spreading to the pure agent-identity layer.
The largest profit pool in AI healthcare sits not at the model layer but at the clinical-workflow entry point, the billable care episode, regulatory compliance, and the high-quality data loop—a layer Epic, Microsoft Dragon Copilot, and Veeva Vault Agents already occupy. The fastest and most certain commercialization is in clinical voice documentation and workflow automation: Microsoft DAX assisted more than 3 million conversations last month across 600 institutions, and Oracle Clinical AI Agent cut physician documentation time by roughly 30% per day. The FDA has authorized over 1,200 AI-enabled devices, yet high-risk diagnostic and therapeutic AI remains tightly regulated. Precision diagnostics and multi-omics form a clear profit pool: Tempus posted 2025 revenue of $1.27 billion (Data Apps $316.4 million, +30.9%), Guardant Shield secured CMS ADLT pricing, Heartflow generated 2025 revenue of $176 million at a 76.8% gross margin, and iRhythm Zio booked Q1 2025 revenue of $158.7 million at a 68.8% gross margin. The most overrated bets: general-purpose medical Agents, autonomous diagnosis, and the idea that AI immediately throws off blockbuster cash flow. Rating Watch: a structurally attractive but slow-validating sector—key names to track are Tempus / Veeva / Microsoft / Epic / Heartflow / iRhythm / Guardant / Exact Sciences / Viz.ai / Aidoc / Recursion / Schrödinger.
AI energy management and the smart grid are not peripheral plumbing around compute; they are the constraint layer that decides whether AI capex can actually turn into deployable capacity. The links with real revenue today are electrical equipment, storage dispatch, DR/VPP, AMI, and utility-digitalization software. This report strongly favors three threads: tier-one suppliers of transformers, switchgear, UPS, and liquid cooling; storage-dispatch software; and grid sensors.
The fastest-monetizing parts of AI in finance are tools that combine high frequency, high pain points, embedded workflows, and auditability: payment fraud prevention (Adyen Uplift, Visa Protect, Mastercard DI), AML/KYC/transaction monitoring, research-document and earnings-call parsing (FactSet, Moody's, LSEG, S&P), credit and lending decisions (FICO, Upstart), wealth-advisor copilots, and fixed-income post-trade processing (Broadridge/LTX). The profit pool is more likely to land in composite platforms that pair data, workflow, and compliance rather than in the pure model layer. JPMorgan's LLM Suite reached 200,000 employees in 8 months and Morgan Stanley's Debrief/AskResearchGPT are at scale, but these remain primarily internal efficiency tools, and new external software revenue will take longer to materialize. Among AI-native names, AlphaSense's ARR tops $500 million, Hebbia's Series B raised $130 million, Quantexa's Series F valued it at $2.6 billion, and Feedzai is valued around $2 billion, yet penetration and pricing power are still unproven. Rating Overweight: the durable winners are the owners of trusted data, embedded workflows, and accountable, auditable outputs.
AI education is a composite value chain of "content rights + learning data + learning workflows + model inference + distribution channels"—the long-run winners are the companies that own authoritative content, assessment systems, school/enterprise entry points, and learning-behavior data, not the pure AI front ends. What commercializes first is not the "universal AI tutor" but products that embed into existing payment structures: Duolingo packs AI into subscription tiers, Pearson embeds AI into Study Prep and assessment, Udemy/Docebo sell enterprise-seat add-ons, and Turnitin makes AI transparency a procurement line item for institutions. The "static answer bank" is the first to be reconstructed by general AI: Chegg's Q2 2025 revenue fell 36% and subscriptions fell 39% (management attributes the decline to Google AI Overviews). AI-native challengers: Speak (Series C of $78 million, valued at $1 billion in late 2024), Preply (Series F of $150 million, valued at $1.2 billion in early 2026), ELSA, Sana, Workera. Key names to track: Duolingo / Pearson / Udemy / Docebo / Coursera / Turnitin / PowerSchool / Instructure / iFlytek / Chegg. Rating Watch: durable profit pools belong to owners of authoritative content, assessment, and channels, not the pure AI front ends.
On-device AI has moved from "feature enhancement" to "the right to define the device" — the real profit pool sits in the OS entry point, the default assistant, the chip platform, the developer API, and the multi-device account system. Microsoft Copilot+ PC, Apple Intelligence, Google Gemini, Amazon Alexa+, and the Meta-EssilorLuxottica smart glasses form five primary fronts, yet only a few have produced evidence that consumers will actually pay.
AI drug discovery has entered tiered commercialization—the first revenue to materialize comes not from "AI inventing new drugs" but from the software, data, simulation, clinical and compliance platforms wrapped around pharma R&D workflows. The FDA received 500+ regulatory submissions containing AI components between 2016 and 2023. Five categories carry clear commercial value: computational chemistry software, biosimulation, clinical trials/RWD, R&D data platforms/ELN/LIMS, and lab automation. The near-to-mid-term profit pool sits with the "pick-and-shovel" vendors and "workflow platforms" (Veeva, IQVIA, Certara, Schrödinger, Tempus, Thermo Fisher, Danaher, 10x, Illumina); the mid-to-long-term upside is reserved for AI-native biotechs that combine platform + pipeline + closed-loop experimentation (Isomorphic, Insilico, Iambic, Generate:Biomedicines, Recursion). Recursion×Sanofi carries up to $5.2 billion in total potential milestones, but what actually lands is the upfront. Insilico has compressed target-to-PCC to roughly 18 months. The frothiest corner: single-point AI protein-design/antibody-design/molecule-generation tools—the structure-prediction moat is weakening (AlphaFold/RoseTTAFold All-Atom are commoditizing structural information). Rating Watch: own the workflow and data layers first, treat pure-model bets with caution.
AI coding has graduated from "code completion" to a "software delivery control plane"—the bottleneck has shifted away from the model toward context, permissions, testing, CI/CD, and governance. Direct revenue evidence: Cursor's annualized revenue topped 2 billion USD by February 2026 and it raised in April at a 50 billion USD valuation; GitHub Copilot has 20 million paying users with Copilot Pro+ up 77% quarter over quarter in FY26 Q2; TCS reports 2.3 billion in annualized AI revenue and HCLTech 620 million in Advanced AI ARR. The most durable winners are not "the model that codes best" but platform entry points (GitHub/GitLab/AWS/Google) plus code-governance layers (JFrog/Datadog) plus modernization migration (IBM watsonx Code Assistant for Z / Amazon Q transformation). Rating Watch: a control-plane shift where platform and governance incumbents, not the flashiest models, capture the durable profit pool.
Data security is shifting from a compliance afterthought into the primary control plane for putting AI into production—Copilot, Azure AI Search, Unity Catalog, Snowflake Horizon, and Bedrock Guardrails all push ACLs, labels, classification, and retrieval authorization upstream into the call chain. RAG and agents do not rebuild the permission system; they amplify the existing over-sharing baked into SharePoint, email, CRM, and databases, with DSPM, DLP, DDR, access governance, and RAG permissions strung into one chain. Machine identities already outnumber human ones 82:1 (CyberArk), and CrowdStrike's $740 million acquisition of SGNL pulls NHI and AI identity into continuous control. Watch rating: the durable profit pool sits in the control plane (permission graph, label engine, policy engine, retrieval authorization, AI gateway) rather than in scanning, with direct beneficiaries backed by financial evidence including Varonis, CyberArk, Snowflake, MongoDB, Elastic, Trend Micro, Cloudflare, CrowdStrike, Palo Alto, Microsoft, and Google Cloud, while AI-native challengers such as Cyera (valued at $9 billion), BigID, Sentra, Concentric, Securiti, Privacera, Veza, Noma, and Lasso carry strong narratives but thinner revenue proof.
In clinical development, AI is no longer a question of whether it works, but of which layer the value lands on. The first revenue does not flow to fully autonomous clinical AI, but to modules embedded in the workflow that connect to real data and remain auditable: patient screening, trial feasibility, site selection, EDC/eCOA/eTMF, risk-based monitoring, RWD/RWE, pharmacovigilance, and partial document automation. The profit pool sits in workflow control (Veeva subscriptions; Medidata/Oracle/Medable platform fees), data access (Tempus, TriNetX, Truveta, Flatiron, Aetion), and compliance capability. AI clearly compresses operational time (Syneos site startup from months to 24-48h; Flatiron 37 seconds per CRF; Medidata listing generation -90%, review -80%; Medable eCOA build from days to 30 minutes), but the evidence that it shortens biological-uncertainty time or lowers failure rates remains weak. The FDA released its AI decision-support guidance framework in 2025, pushing the sector into a phase of dual regulatory and commercial validation. Rating Overweight: a value chain where data uniqueness, workflow depth, compliance, and clean billing decide who captures the durable profit pool.
AI content and creativity has moved from "model demos" into "workflow commercialization," and the profit pool flows mainly to platform companies: Adobe posted $23.769 billion in FY25 revenue, $19.2 billion in Digital Media ARR, and over $250 million in Firefly ARR by Q1 FY26; Canva and Figma have embedded AI into the design OS; in a single quarter Meta saw over 1 million advertisers use GenAI tools to produce 15 million+ ads; Kuaishou's Kling booked RMB 340 million in 2025Q4 while AIGC marketing assets drove RMB 4 billion of online marketing spend; Meitu's 2025 imaging/video/design revenue reached RMB 2.954 billion with 16.91 million paying users. Copyright and training-data licensing is rising into a new revenue line: Getty and Shutterstock lead with commercial-use rights plus indemnity plus licensed data, UMG and WMG are shifting from "suing AI" to "conditional licensing," and Disney/Universal/Warner are suing Midjourney while OpenAI has retired its old Sora. Still in pilot: feature-length video generation, open AI-music platforms, standalone consumer AI apps, and AI virtual idols. Top risks: copyright litigation, inference cost and price wars, and platform free-tier substitution. Rating Overweight: the durable winners are integrated platforms that own distribution, workflow, budget access, and copyright governance, not point-solution generators.